Distributed Wireless Body-Centric Networks: Optimisation via Predictive Analytics and Cross-Layer Designs

Abstract

In this thesis, methods for optimisation are designed and developed for wireless body-centric channels in order to best enable intelligent, self-organised, and distributed human-centered networks. Due to human-body shadowing and mobility, body-centric channels change dramatically over time --- with intermittent periods of longer stability --- often causing unnecessary delay and energy consumption under significant radio channel attenuation. Hence, robust and efficient communications across these networks require adaptive, optimised mechanisms. Thus, we propose, and investigate the performance of various cross-layer and predictive optimisation techniques for real-life body-centric channels. Our analysis employs real-life experimental datasets collected from different numbers of co-located wireless body area networks (BANs), over many hours in 'everyday' scenarios.

We first investigate cooperative receive diversity for a BAN used to monitor a sleeping person, and we find that cooperative combining improves packet delivery ratio (PDR) and latency for these atypical slowly-varying radio channels. Then, we propose two cross-layer optimised techniques for multiple coexisting BANs, forming wireless body-to-body networks (BBNs). These techniques are shortest path routing (SPR) and cooperative multi-path routing (CMR), where CMR incorporates cooperative selection combining. In CMR and SPR, the best route is periodically selected at the network layer according to channel state information from the physical layer. We show that CMR reduces retransmissions and increases PDR due to an available alternate path, reducing end-to-end delay and energy consumption with respect to state-of-the-art protocols. We then add MAC layer interference mitigation using low duty cycle TDMA, for which CMR gains up to 14 dB improvement over SPR at 10\% outage probability. Moreover, we also apply CMR incorporating novel MAC layer CSMA/CA with adaptive carrier sensing, giving improvement over TDMA for both throughput and spectral efficiency.

Next, we explore the feasibility of applying predictive optimisation over wireless body-centric channels (i.e., body-to-body, on-body), by studying wide-sense-stationarity (WSS) and long-range dependence (LRD) characteristics, which are crucial to predictive analysis. The results of different stationarity tests show that unlike on-body channels (which are considered non-stationary), body-to-body channels possess WSS characteristics for a range of window lengths between 0.5 s and 15 s (typically 5-8 s) depending upon on-body sensor locations and shadowing. Moreover, the Hurst exponent of body-centric channels is very high (around 0.9) and temporal auto-correlation decreases very slowly (a power-like decay), indicating LRD (i.e., long-memory) characteristics are maintained.

Then, based on the results for WSS and LRD, we apply multi-objective optimisation for adaptive scheduling in BBNs, by using a multi-objective Markov decision process (MOMDP) to jointly optimise three separate metrics: throughput, latency, and energy consumption. The adaptive scheduling combines both TDMA and CSMA/CA schemes. From performance analysis, employing real-life channel measurements, we find an MOMDP outcome that is Pareto optimal, providing a desirable trade-off between the three objectives of maximising throughput, and minimising continuous latency and energy consumption. It is also observed that WSS characteristics of the body-to-body channels have a significant effect on the outcome of such analytics.

The outcomes here help the development of pervasive real-world applications with large-scale and highly-connected systems, comprising many closely-located BANs.